AI That Learns You Over Time

Apps like So Syncd match you on MBTI. Others use DISC, the Big Five, or attachment style questionnaires. You take a test once. Fill it out in ten minutes. The algorithm uses it to match you indefinitely.

But you're not the same person you were when you took that test. Your priorities shift. Your context changes. You move cities. You start a business. You finish a hard year. What you need from another person at 27 is different from what you needed at 23, or what you'll need at 34.

A static personality type can't capture any of that.

The problem with one-time personality tests

The research on MBTI reliability is not kind to it. Studies have found that up to 50% of people who retake the MBTI within five weeks get a different four-letter result. The same person, five weeks later, categorized differently by the same instrument.

This isn't a minor measurement error. It's evidence that the categories themselves are unstable — that personality is more fluid than the test assumes, and that the test-retest reliability of MBTI is poor compared to what you'd expect from a scientific measurement tool.

The Big Five (OCEAN) is more scientifically grounded, with better test-retest reliability. But even a stable Big Five profile doesn't tell you much about what kind of connection you're looking for right now. You can be high in openness and be a homebody this month because you're deep in a project. You can be introverted but want a training partner who pushes your pace.

Personality type is a rough sketch of a person, not a map.

What "learning AI" actually means

There are two models for how an AI matching system can work.

Static model: You take a test or fill out a profile. The system extracts a fixed set of attributes — type, interests, demographics. Every future match is derived from that fixed representation of you. You change; the system doesn't notice.

Adaptive model: The system starts with your initial profile but updates its understanding based on feedback loops. How did that match go? What did you actually talk about? What worked and what didn't? Future matches incorporate that signal.

Most personality-test-based apps use the static model. They have to — the test is the product. If the test result could change, the whole premise of "you're an ENFP, here are your compatible types" falls apart.

An adaptive model is harder to build. But it's the only one that actually tracks you as a person rather than a snapshot.

Why context matters more than type

What you need from a connection depends heavily on what you're doing.

You need different people for different roles: a running partner who shows up at 6am regardless of weather. A mentor who's three years ahead of you in your industry. Someone to date who matches your current bandwidth for a relationship — which, depending on the year, could be "a lot" or "very little." A co-founder who can work alongside you without friction.

A single personality type label can't describe all of those needs at once. And even if it could, the same type might fit a completely different person in each role.

Context is what actually determines who you need right now. Not your Big Five scores.

The personality test trap

Personality type tests are genuinely fun. They're shareable. "I'm an INTJ, what are you?" is a usable conversation starter in a way that "I scored 68th percentile on conscientiousness" is not.

That social utility is real. But it doesn't mean the tests are good matching instruments.

The research on whether MBTI type predicts relationship quality or longevity is weak. Couples of "compatible" types don't report meaningfully better outcomes than couples of "incompatible" types. Shared activities, proximity, and aligned goals — these have stronger predictive power.

So Syncd is building a product around the weaker signal and calling it science. It's entertaining. It probably helps some people. But it's not the best tool for finding someone worth knowing.

The strongest predictors of connection and relationship quality are things like shared values, compatible life goals, geographic proximity, and mutual investment in the relationship. Those are contextual, situational, and specific. They don't reduce to four letters.

How Sphere approaches learning

Sphere doesn't ask you to take a personality test. It asks you what you're doing right now, what you're working toward, what your schedule actually looks like, and what kind of connection you're looking for.

Those answers are much closer to what actually matters in a match.

And they change. When your situation changes, you update your profile. When a match leads to a connection, that signal informs future matching. The system isn't starting from a fixed type — it's building a current picture of where you are and what you need.

It's situational, not typological. That's a deliberate design choice.

The bottom line

"Who are you?" is a reasonable question. But it's less useful for matching than "what do you need right now?"

Personality type tests answer the first question, slowly and imprecisely. They take a snapshot and treat it as identity.

What actually matters for finding the right person is contextual: your current goals, your schedule, your bandwidth, your life stage, the specific kind of connection you're looking for. Those things evolve. A good matching system evolves with them.

That's the premise Sphere is built on. Not "what type are you" but "what do you need, and who has that to offer right now?"

See also: best Tinder alternatives in 2026 and Sphere vs Sitch.